compare_subjectsRank up to 50 subjects on any criteria.

When an agent has to choose — between candidates, between vendors, between potential acquisition targets, between three Series-A startups in the same category — what it actually needs is a side-by-side matrix. compare_subjects takes a list of up to 50 subjects and a free-text criteria description, runs the same scoring pass over each one, and returns a deduplicated ranked list with the underlying matrix attached. One call. Ten credits. The agent gets a defendable answer instead of three half-formed ones.

When to call this tool

The tool is purpose-built for one of the most common agent shapes: "given these N candidates, which is best, and why?" That shape shows up in recruiting (rank 12 finalists), in vendor selection (rank 8 SaaS tools on the same criteria), in journalism (rank 20 climate-tech founders by recent activity), in venture (rank 15 startups in a category by signal density). Without this tool, an agent does it the slow way: one read per subject, then a synthesis pass, then a justification pass. With it, the agent gets all three in one shot, with consistent scoring across subjects.

The criteria string is interpreted by the tool — you write the criteria the way you'd brief a junior analyst, and the tool maps it to the right scoring axes. "Senior engineering leaders with public-speaking presence and shipped infra at scale" becomes three measurable axes; "growth-stage SaaS in the developer tooling space with founder-led marketing" becomes three different ones.

Input schema

{
  "subjects": ["string  // 2-50 names, handles, or domains"],
  "subject_kind": "enum: 'person' | 'company' | 'product' | 'auto'",
  "criteria": "string  // free-text, mapped to scoring axes",
  "weight_hints": "object  // optional, per-axis weights 0-1",
  "include_matrix": "boolean  // default true"
}

Output schema

{
  "axes": [{ "name", "weight", "description" }],
  "matrix": [{ "subject", "axis_scores": { "axis_name": 0-100 } }],
  "ranked_list": [{ "rank", "subject", "composite_score", "why" }],
  "sources": ["url"],
  "credits_used": 10
}

Example invocations

1. Claude Desktop

// User prompt
Compare Anthropic, OpenAI, and Google DeepMind as employers
for a senior systems engineer who values shipping over PR.

2. ChatGPT custom GPT

// GPT instructions
When the user gives you a list of options and asks "which is
best", call compare_subjects with their list and their stated
criteria. Always show the ranked_list with reasons before any
recommendation.

3. Cursor MCP

// In Cursor
Use compare_subjects to rank these 12 OSS metrics tools for
a small SaaS team. Output a markdown table.

4. n8n (HTTP Request fallback)

{
  "jsonrpc": "2.0", "id": 1, "method": "tools/call",
  "params": {
    "name": "compare_subjects",
    "arguments": {
      "subjects": ["Datadog", "Grafana", "New Relic"],
      "criteria": "open-source friendliness, total cost of ownership at 50 hosts"
    }
  }
}

5. Raw curl

curl -X POST https://www.mentionfox.com/mcp \
  -H "Authorization: Bearer $FOXAPIS_KEY" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"compare_subjects","arguments":{"subjects":["Stripe","Adyen","Checkout.com"],"criteria":"developer experience and Asia coverage"}}}'

Sample output (real, redacted)

{
  "axes": [
    { "name": "developer experience", "weight": 0.5 },
    { "name": "asia coverage", "weight": 0.5 }
  ],
  "ranked_list": [
    { "rank": 1, "subject": "Stripe", "composite_score": 87,
      "why": "Strongest documentation and SDK breadth. Solid Asia presence in JP/SG/IN." },
    { "rank": 2, "subject": "Adyen", "composite_score": 79 },
    { "rank": 3, "subject": "Checkout.com", "composite_score": 71 }
  ],
  "credits_used": 10
}

Credit cost & rate limits

Flat 10 credits for any list size from 2 to 50 subjects. Rate limit is 20 calls per minute per key. The list size is hard-capped at 50; pass more and the call returns 422 LIST_TOO_LARGE. For deeper per-subject reads, chain compare_subjects with vet_person on the top 3 ranked items.

Error codes & recovery